Well, CRISPR is more flexible in targeting a new sequence, but the current crop of therapies already have screened TALEN / ZFN mutants that target their gene of interest, so that advantage is somewhat moot.
The non-specificity is I think a much bigger problem than you expect due to the fact that Cas9 can cut sites where it has at least a partial match of its guide RNA to the DNA. Even if there is only one exact match to the gRNA in a genome, there are still millions of PAMs where Cas9 will at least transiently bind, and a partial match in the early sequence can cause the enzyme to stick around long enough to cleave the DNA in a non-trivial fraction of the time.
Keep in mind that Cas9 was primarily evolved to target viral genomes, which are generally pretty small (well less than 1 Mb) in a background of bacteria genomes (~ 2 Mb, say, for
S. pyogenes). In contrast, the human genome is about 3,000 Mb in size, which means you need 3 orders of magnitude more specificity than the protein was evolved for. Not impossible, but it means you need a bit of tweaking to the system.
There are several tricks people have come up with to reduce this off-target cleavage, but it's still not a solved problem to the point that you would feel safe putting the system in a life human body. Add to that the on-going patent dispute over Cas9 and you have a pretty good incentive to wait it out while academic researchers iron out the bugs with the technology.